Measuring output impedance

[ruffrecords]

Most pro gear will drive a 10K load and will also drive a 1K load. With the indirect method you send a tone(typically 1KHz at 0dBu) to the output and measure the actual value of the output, once with a 10K load and then with a 1K load. I have chosen these values because the considerably simplify the maths. Divide the 10K voltage by the 1K voltage level and call this ratio X.  If my math is correct, the output impedance (Rout) in kilohms is:

Rout = 10(X-1)/(10-X)

Cheers

Ian

 

[johnheath]

Thank you Ian

I will try that tomorrow.


Best regards

/John

 

[abbey road d enfer]

I am sorry but Rg is not understood for me here  :-\

It's the source impedance of the oscillator itself.
The "g" index comes from "generator"; alternatively, one often uses Rs, where the "s" index is for "source".

 

[PRR]

> send a tone (typically 1KHz at 0dBu) to the output and measure the actual value of the output

Working back, it appears that you set the reference level NO-load?

Actual level not critical, except it must be well below overload yet high enough to read with desired precision.

> Divide the 10K voltage by the 1K voltage level and call this ratio X. 
> the output impedance (Rout) in kilohms is: Rout = 10(X-1)/(10-X)

Let us write Rout = 10*((X-1)/(10-X))

Parenthesis reduces ambigulation for those of us who forget order of operators.

Worked example:

1K 1V no load source
Loaded with 10K: 0.909V
Loaded with 1K: 0.5V
X= 0.909/0.5= 1.818
X-1= 0.818
10-X= 8.18
(X-1)/(10-X)= 0.818/8.18= 0.1
10*0.1= 1.0

> (Rout) in kilohms

Actually in the value of the lower load. 1K is indeed safe and convenient for most "line" outputs. In an older shop they might have a 6K and a 600, find the result in multiples of 600r. "Loose" speaker amps (low DF) could use 80 and 8.

Abbey will correctly object that there ARE impedance correcting and negative impedance outputs. I personally believe these should be scrapped as more mysterious trouble than any possible benefit. But people here insist on some strange circuits. They may give dubious results (on test as in practice).

 

[ruffrecords]

> send a tone (typically 1KHz at 0dBu) to the output and measure the actual value of the output

Working back, it appears that you set the reference level NO-load?

Yes and no.  Unless you have the schematic to hand, you are never quite sure if there really is no load, but it does not matter anyway. The important thing is that you don't change anything between measurements other than the load. essentially the assumption is the output Z is purely resistive and we can therefore plot the straight line load line from two points. In most cases this is more than sufficient.

Abbey will correctly object that there ARE impedance correcting and negative impedance outputs. I personally believe these should be scrapped as more mysterious trouble than any possible benefit. But people here insist on some strange circuits. They may give dubious results (on test as in practice).

I am uncomfortable with this. If such an output cannot drive both a 1K and 10K loads it seems a bit of a retrospective step. If it can, the implication is it can adjust its output impedance depending on the load in which case measuring its output impedance seems rather pointless.

Cheers

Ian

 

[abbey road d enfer]

I am uncomfortable with this. If such an output cannot drive both a 1K and 10K loads it seems a bit of a retrospective step. If it can, the implication is it can adjust its output impedance depending on the load in which case measuring its output impedance seems rather pointless.

My point was, more generally speaking, that changing significantly the load changes the behaviour, in a way that may not be subtle.
That is a well-known fact; the basic method for simulating open-loop gain and all other relevant parameters was traditonally to alter the NFB network with a large inductor and a large capacitor that would allow proper DC analysis before doing the AC analysis. It has been shown that it modified the operation of the circuit, because the impedances of the nodes are not anymore what they are in normal operation. As Ricardo mentioned a few posts earlier, what the circuit is connected to matters.
Many tube amps need to see a load that's pretty close to nominal to operate correctly; it does not make them particularly quirky.
Just think of atube amp that's designed for a nominal 4 ohms load; when you load it with 16 ohms, it's close-loop gain suddenly jumps by about 6 dB, but global NFB maintains gain at almost the same, which results in actually halving the output Z! That's why an impedance measurement cannot be taken in ignorance of the load it is supposed to be attached to.
Now you may find that a McIntosh is much less sensitive to its environment than a Dynaco, but it doesn't make the latter a product to reject shamefully.
Indeed, negative-impedance drive and distortion cancellation schemes that rely on too much NFB can create havoc, however, there are many existing products that testify to the fact that, when well-designed, they can significanty improve performance without any drawbacks.
Many of my commercial products used a tertiary winding NFB scheme that improved performance in many respects:
LF response: -0.1dB @20Hz instead of -1 for the bare xfmr
Output Z: 22r instead of 44
LF THD: 0.04% @ 22Hz +20 dBu instead of nearly 1%!
All measured into 600r
Susceptibility to external magnetic fields reduced by 30 dB (big saving here since it saves the costs of a magnetic shield)
And not a single issue as to stability.

 

[ruffrecords]

@Abbey,

I am happy that there are some particular amplifiers, especially some tube power amplifiers, that are designed to work into a single specific load impedance. Not surprisingly, their characteristics change if you don't use the specific load impedance. Clearly, using the indirect method in those cases is inappropriate.

There are also lots of amplifiers that use straightforward NFB (including or excluding any transformer) that provide all the benefits you get from NFB and which can drive a wide range of loads (10K down to 600 ohms). The indirect method is fine for these.

Then there are some amplifiers that also use positive feedback (PFB), often including a transformer, to allow a smaller, cheaper, lower inductance transformer to achieve the performance that would require a larger transformer without the PFB. These are the ones that I an doubtful about. PFB is common in audio electronics, bootstrapping for example. but it is usually confined within a circuit and is not exposed to te direct influence of the outside world. PFB with output transformers does however expose the PFB network to the vagaries of the outside world and that I do not fully understand,

Cheers

Ian

 

[abbey road d enfer]

@Abbey,

I am happy that there are some particular amplifiers, especially some tube power amplifiers, that are designed to work into a single specific load impedance. Not surprisingly, their characteristics change if you don't use the specific load impedance.

Are you being sarcastic here? Almost any amplifier sees its instrinsic performance change with the load it sees. This is evidence.
What is not so evident is that an amplifier with NFB also sees its characterisics change with load; the output impedance in particular changes, for the reason I wrote before. But it's true that the level and frequency response change very little.

Clearly, using the indirect method in those cases is inappropriate.

The reasons for using one or the other method are not because of variable impedance; both methods would show the same results, as long as they are accurately implemented. Most of the times, the difference is so subtle it takes extreme metrologic precision to detect them. I mentioned earlier the necessity of making sure that the measurement set-up made the operation as close as possible to nominal use. Clearly, the direct method with a current-source (infinite source impedance) does not confirm this. It is more correct to use a low impedance source in series with resistors close to the nominal load.

There are also lots of amplifiers that use straightforward NFB (including or excluding any transformer) that provide all the benefits you get from NFB and which can drive a wide range of loads (10K down to 600 ohms).

The fact that they can drive a range of loads does not contradict the fact that some characteristics change. NFB is not an eraser.  Since almost inevitably, a tube power amp must have an OT, the designer builds the amp for a singe operation point; he does not need to cater for variable load, he just assumes the proper tap will be useed. It doesn't mean that they wouldn't make sound when improperly matched to their load.
OTOH, "modern" tube line amps are more tolerant by design (the designers knows it must be capable of driving a range of loads). Vintage units were designed for strict impedance matching, though.

  Then there are some amplifiers that also use positive feedback (PFB), often including a transformer, to allow a smaller, cheaper, lower inductance transformer to achieve the performance that would require a larger transformer without the PFB.

They are also used to achieve stellar performance by pushing up a couple of notches the performance of already excellent transformers. Again, the AP oscillator is a paragon of this technique, that has also been extensively used by Studer and NTP, that are hardly second-rank manufacturers.

  These are the ones that I an doubtful about. PFB is common in audio electronics, bootstrapping for example. but it is usually confined within a circuit and is not exposed to te direct influence of the outside world. PFB with output transformers does however expose the PFB network to the vagaries of the outside world...

Then you could adress the very same criticism at NFB. The load affects more or less directly the open-loop gain, thus the NFB ratio. Indeed, some PFB implementations are more sensitive to the external world than others; it is the responsibility of the designer to make sure his design is adequate.

  ...and that I do not fully understand,

  Just think; how much is a tertiary winding exposed to the outside world?

 

[CJ]

maybe plot distortion vs load

 

[moamps]

The output impedance of preamps, opamps etc. can be easily measured using direct method and any good software for speaker's impedance measurement. Here is the graph of 2520 opamp output impedance  for two gains (made in hurry) . ARTA (Limp module ) software is used.

 

[ruffrecords]

Are you being sarcastic here?

Not at all. I am genuinely interested in improving my knowledge. I apologise if the way I worded it sounded sarcastic.

Cheers

Ian

 

[abbey road d enfer]

The output impedance of preamps, opamps etc. can be easily measured using direct method and any good software for speaker's impedance measurement. Here is the graph of 2520 opamp output impedance  for two gains (made in hurry) . ARTA (Limp module ) software is used.

Impedance measurement requires only a generator and a voltmeter (possibly a phase-meter).
The whole issue there is the set-up. I suspect you have used the ARTA recommende set-up, with a resistor in series with the load and measurements being taken on both sides of the resistor. What value of resistor have you used? Have you done successive measurements with different values of this resistor?
The measured output Z of 0.12-0.15 ohm implies that any sensible load will look like open-circuit to the opamp.
That is generally the case with most SS transformerless designs.
The difficulty is that most vacuum tube circuits (and some SS too) are different in that respect. Their open-loop varies with load, and since global NFB stabilizes closeed-loop gain, the resulting closed-loop output Z varies; that's why I suggest, for these cases, and whatever the chosen method, using loads as close as possible to nominal.
In the case of the indirect method, which involves submitting the DUT to varying loads, that implies choosing loads that are not too distant from nominal like half-load, nominal load and double-load, but that implies very careful optimization of reading errors.
The direct method is intrinsically better, and less prone to reading errors. It is however sometimes a tittle awkward to implement, particularly in the case of electronically balanced output stages.
Anyway, for what it matters in most cases (typically knowing if the output is capable of driving 10k or 600), the indirect method is perfectly adequate (but I wouldn't use the results in a documentation for a government agency).

 

[abbey road d enfer]

I apologise if the way I worded it sounded sarcastic.

Chalk it up to English not being my native language.

 

[moamps]

Impedance measurement requires only a generator and a voltmeter (possibly a phase-meter).

And a calculator, plenty of time etc.....  Here PC with audio card and few high precision resistors can be used for very quick and accurate measurements. 

The whole issue there is the set-up. I suspect you have used the ARTA recommende set-up, with a resistor in series with the load and measurements being taken on both sides of the resistor.

That's the setup for speaker's impedance measurement, isn't?

What value of resistor have you used?

6 ohms in this case.

Have you done successive measurements with different values of this resistor?

No, the result should be the same in this case (discrete SS preamp with lot of NFB and output current available). The result changes more with different NFB applied.

The difficulty is that most vacuum tube circuits (and some SS too) are different in that respect. Their open-loop varies with load, and since global NFB stabilizes closeed-loop gain, the resulting closed-loop output Z varies; that's why I suggest, for these cases, and whatever the chosen method, using loads as close as possible to nominal.

The real question to me is why someone will like to  measure output impedance if there is strictly defined load impedance ?

In the case of the indirect method, which involves submitting the DUT to varying loads, that implies choosing loads that are not too distant from nominal like half-load, nominal load and double-load, but that implies very careful optimization of reading errors.

We can use here tenth deviation of nominal load, up and down, simultaneously checking THD and IMD, and plot very easily difference in output voltage and corresponding output impedance.

The direct method is intrinsically better, and less prone to reading errors. It is however sometimes a tittle awkward to implement, particularly in the case of electronically balanced output stages.

Agreed. We should always know how and why our measurement method can introduce errors in real working conditions of a DUT.

 

[abbey road d enfer]

The whole issue there is the set-up. I suspect you have used the ARTA recommende set-up, with a resistor in series with the load and measurements being taken on both sides of the resistor.

That's the setup for speaker's impedance measurement, isn't? 

Yes, isn't it the one you've used?

What value of resistor have you used?

6 ohms in this case. 

So you drove a 6 ohms load from the output of your souncard? Or did you have some kind of buffer/booster?

The real question to me is why someone will like to  measure output impedance if there is strictly defined load impedance ? 

No, not strictly defined, but you know, there are specs. Typically, they would read as "recommended load impedance: 600 ohms or more", but they don't say how load reacts with performance and how precise is 600 ohms, is it +/- 10% or +/- 1ppm?  I know it is irrelevant in practice, but sometimes numbers are used in a pass/no pass situation.

We can use here tenth deviation of nominal load, up and down, simultaneously checking THD and IMD, and plot very easily difference in output voltage and corresponding output impedance. 

In a typical case of a DUT with an output Z of 50r and nominal load Z of 600, that would mean measuring with a 545r a 600 and a 660r. In these conditions, the voltage sag is ca. 0.13 dB.
A variation of 10% in the output impedance results in a deviation of ca. 0.01dB of the sag measurement. Detecting a lesser variation in output impedance takes an unusual combination of instrument accuracy, knowledge and skills.
That is why I claim this method as much less accurate than the direct method, whenever (if ever) accuracy is required.

 

[johnheath]

It's the source impedance of the oscillator itself.
The "g" index comes from "generator"; alternatively, one often uses Rs, where the "s" index is for "source".

Ah - thank you sir… I was confused there for a while

Best regards

/John

 

[johnheath]

Actually in the value of the lower load. 1K is indeed safe and convenient for most "line" outputs. In an older shop they might have a 6K and a 600, find the result in multiples of 600r. "Loose" speaker amps (low DF) could use 80 and 8.

Abbey will correctly object that there ARE impedance correcting and negative impedance outputs. I personally believe these should be scrapped as more mysterious trouble than any possible benefit. But people here insist on some strange circuits. They may give dubious results (on test as in practice).

Thank you sir

So, if I understand the content of the discussion most modern devices have a 10k input load and many times a 50R source. So if you have a device with a source of 1k it would stay within the 1:10 ratio for proper drive capability to a device of a 10k load?

And for the same ratio driving a 600R load, 50R (60R) is needed?

Best regards

/John

 

[johnheath]

Thank you all

It has been very interesting to follow your comments back and forth… unfortunately a lot of it is way above my knowledge.

But what I have picked up so far is that the "direct" method is "better" for measuring the output impedance of line level tube devices.

I attached a hasty sketch of how I interpreted how I could hook it all up but I am still not sure if that is the proper way. Abbey suggested that a voltage AC, a given resistor and a scope would be the ingredients of the "direct" method.

Ian suggested to measure with two loads 1k and 10k and also added a formula for the math which seems pretty simple to handle.

Moamps gave me an interesting idea if I could make a device on its own… like…just plug and measure?

Now I really could use some guidance on how to hook it all up… or is my sketch the way to do it?

ps. Sorry if it all is implied somewhere in your comments and I have missed it.

Best regards

/John

 

[Audio1Man]

Hi John
You question about measuring impedance has had many good answers. I use an Audio Precision and some other equipment to measure impedance. This will give a graph of the impedance covering 10Hz to 200kHz.  I make a current source and drive the “LOAD”. This plots the impedance vs frequency. Now replace the load with a fixed known resistor and plot it again. His will calibrate the graph.
The current source can be the AP with a large value fixed resistor (10x to 100x the value of Z). If you need a big voltage I then use a power amplifier and a big resistor. If the frequency response of the power amplifier is not flat it can be corrected by EQ added to the AP generator.
This method can be used with several other instruments, however some may not have all of the same functions.
Another is use an audio generator, current source, ac meter, loads, pencil & paper or Excell file.

Duke

 

[moamps]

So you drove a 6 ohms load from the output of your souncard? Or did you have some kind of buffer/booster?

I use 15W IC power amp for speakers measurement at different membrane excursion. So I used it here as is just to show how someone can easily and fast measure impedance and phase. The output impedance of this opamp is very low so it can be done this way. If high output impedance of a DUT is expected, the sensing resistor can be higher and direct line/headphone output from a soundcard can be used. 

In a typical case of a DUT with an output Z of 50r and nominal load Z of 600, that would mean measuring with a 545r a 600 and a 660r. In these conditions, the voltage sag is ca. 0.13 dB.

Do you mean that difference in voltage sags is 0.13dB?

For Zg=50 and Zl=545, 600 and 660, voltage sags are 0.76dB, 0.69dB and 0.63dB.
If the output voltage from the DUT is for example 1V, the measurement deviation from Zl=545 to 660 is  13mV what's 38dB below 1V and it is within measurement range of any decent soundcard.